The joint probability density function (PDF) of thermochemical variables can be used for accurately computing the combustion source term in turbulent flows. Evolution of the PDF in physical and composition spaces is governed by a transport equation which requires modeling of sub-filter mixing process. Mixing models used in PDF methods are known to have a substantial impact on reactions, and can therefore signifiantly influence flame stabilization in turbulent combustors. A systematic approach is required for studying the effect of variations in the mixing model constant on reactions. The objective of this work is to develop a sensitivity based frame-work to study this problem. The approach presented here is developed in the context of the direct quadrature method of moments (DQMOM), which is an Eulerian method for solving the PDF transport equation. Governing equations for the corresponding differential sensitivity analysis are formulated and implemented in a practical large eddy simulation (LES) solver. The DQMOM-sensitivity approach is then used for studying an experimental supersonic combustor.